R0053/2026-03-31-02/C002/H2¶
Statement¶
The claim is partially correct — AI does treat weakly-stated requirements as suggestions (diagnosis correct), but the specific mechanism of negative constraints is not the best approach (prescription wrong).
Status¶
Current: Supported
Supporting Evidence¶
| Evidence | Summary |
|---|---|
| SRC01-E01 | Requirements need enforcement but negative framing is suboptimal |
| SRC01-E02 | Positive reframing resolves issues negative constraints cannot |
| SRC02-E01 | Instruction hierarchies fail, confirming enforcement gap |
Contradicting Evidence¶
| Evidence | Summary |
|---|---|
| None | No evidence contradicts the partial-correctness position |
Reasoning¶
All evidence converges: the problem is real (AI ignores loosely-stated requirements) but the solution needs refinement (positive enforcement outperforms negative constraints in most contexts).
Relationship to Other Hypotheses¶
H2 accepts H1's diagnosis but rejects the mechanism. H2 and H3 are contradictory on the diagnosis.